Skip to content
This repository has been archived by the owner on Jun 2, 2024. It is now read-only.

Latest commit

 

History

History

discord

README.md

GPT Turbo - Implementation - Discord

GPT Turbo - Discord License: MIT

Discord bot powered by the GPT Turbo library. Inspired by Discord's very own Clyde AI chat bot.

Installation

Run these at the mono-repo root. Not this package directory

# Installs mono-repo dependencies, bootstraps packages, and builds packages
npm install

Run these in this package directory

# Register slash commands
npm run register

Prerequisites

  • You'll need to setup a Discord bot. You can do this by following the Discord.js guide.
  • Before running the bot You'll need to specify your Discord bot token and client ID in the environment variables. You can do this by creating a .env file in this directory with the DISCORD_TOKEN and DISCORD_CLIENT_ID variables, by setting them in your shell, or by setting them at runtime (see below).

Running the bot

If you plan on deploying this bot for yourself, you'll need to set the following permissions: 397284551744: https://discord.com/api/oauth2/authorize?permissions=397284551744&scope=bot%20applications.commands&client_id=`DISCORD_CLIENT_ID`

Run these in this package directory

# Start the bot
npm start

# Start the bot in development mode
npm run dev

# Start the bot and specify the token and client ID
DISCORD_TOKEN=your-token-here DISCORD_CLIENT_ID=your-client-id-here npm start
DISCORD_TOKEN=your-token-here DISCORD_CLIENT_ID=your-client-id-here npm run dev

Whitelisting / Blacklisting

OpenAI has a gray rule on Bring-Your-Own-Key (BYOK) apps. This means that it's not explicitly clear in their usage policies whether or not you're allowed to ask users of you product for their API keys to use with your app. However, some old forum replies from OpenAI staff suggests that it's not allowed.

It's different for other implementations of GPT Turbo, as they don't store the API key anywhere else than on the user's machine. (source) This implementation would need to store the API key somewhere other than your machine in order to use it for conversations.

That being said, you'll need to provide your own API key for everyone to use. Because you don't want everyone making it paid calls to OpenAI, I implemented a whitelist and blacklist system for the bot. This means that you can specify a list of users/guilds that are allowed to use the bot, and a list of users/guilds that are not allowed to use the bot.

This is done through environment variables. Refer to the .env.example file for more information on how to set this up.

No matter what the whitelist says, the blacklist will always take precedence.

Quota System

The Quota System is opt-in. Specifying a database connection string in your environment variables will enable it automatically. See the .env.example file for more information.

On top of whitelisting and blacklisting, you can optionally enable the bot's quota system. This system allows you to specify a maximum amount of tokens that can be used for each allowed user or role. It uses key-value storage powered by KeyV, which is why multiple databases are supported (MongoDB, PostgreSQL and MySQL).

If you haven't explicitly set a quota for a user or one of the user's roles, the bot will use the default quota of 5000 tokens (about $0.01). This can be changed by setting the DEFAULT_QUOTA environment variable.

Important things to note about the quota system:

  • The usages are never reset automatically. Bot ADMINS (see .env.example) can reset the usages for a user/everyone by using the admin slash command.
  • Bot ADMINS can set a per-user quota by using the admin slash command.
  • The tokens are calculated by gpt-turbo using third-party libraries. The calculated tokens may slightly differ from OpenAI's actual token count. This shouldn't be too much of an issue for small quotas (e.g. 5000 tokens), but it may be an issue for larger quotas (e.g. 1000000 tokens).
  • The bot will set the max_tokens accordingly to prevent the user from going over their quota.
    • For example: A user's usage is at 5/10 tokens and their current conversation size is 2 tokens, this means whatever prompt is next will incur a cost of at least 5 + 2 = 7 tokens. The bot will set the max_tokens parameter to 10 - 7 = 3 tokens to prevent the user from going over their quota. Again, depending on how OpenAI actually respects this parameter, there may be a slight difference in the actual token count.
  • User quotas have priority over role quotas.

The MESSAGE_CONTENT intent

The MESSAGE_CONTENT intent is a privileged Discord bot intent that allows bots to read messages that do not directly involve them (i.e. where the bot is not mentioned or being replied to with ping enabled). Developers can enable it at any time on the Discord Developer Portal, but bots with over 100 servers need to be approved by Discord to use it.

Fortunately, this bot can work with both the MESSAGE_CONTENT intent enabled and disabled. If you have it enabled, you'll need to set the USE_MESSAGE_CONTENT_INTENT=true environment variable for the bot to use the intent (on top of enabling it in the Discord Developer Portal). However, keep in mind the bot was designed to work with it disabled first. You may report any issues you encounter with it enabled.

There are no intent-related limitations in DMs. In DMs, the bot will always create a new conversation when the message is not replying to the bot's previous message.

There are some differences in how the bot works depending on whether or not the intent is enabled:

  • When disabled, the user needs to consistently reply to the bot's messages with ping enabled in order for it to work.
    • Users need to mention the bot in order to start a conversation.
    • Users need to reply to the bot's initial response with ping enabled in order to continue the conversation in a thread.
    • Users need to reply to the bot's messages with ping enabled in a thread (created by the bot) in order to continue the conversation in the thread.
  • When enabled, the user doesn't need to reply to the bot'
    • Users need to mention the bot in order to start a conversation.
    • Users need to reply to the bot's initial response in order to continue the conversation in a thread.
    • Users can send messages in the thread, without the need to reply to the bot's previous message, to continue the conversation in the thread